The dataset comprises labeled genetic variations of the BRCA1 and BRCA2 genes obtained from the ClinVar database. After refining annotations based on quality and completeness (SPDI and pathogenicity fields), artificially generated mutated DNA sequences were created using the reference genome GRCh38. Mutations were introduced to segments (300bp or 500bp) at designated locations, and sequences were categorized as 0 (benign) or 1 (pathogenic) according to annotations. Consequently, sequences that had high-quality review statuses (e.g., expert panel, multiple reviewers) were chosen for fine-tuning the DNABERT2 pre-trained model. To handle class imbalance, extra benign sequences were obtained from the wild-type reference employing a sliding window method. The final dataset contained 4,213 sequences for BRCA1 and 5,996 for BRCA2, further balanced to 5,397 and 6,701 samples, respectively.